DOCUMENT DE TRAVAIL
DT/2016-12
Does Managerial Capital also
Matter Among Micro and Small
Firms in Developing Countries?
Axel DEMENET
UMR DIAL 225
Place du Maréchal de Lattre de Tassigny 75775 • Paris •Tél. (33) 01 44 05 45 42 • Fax (33) 01 44 05 45 45
• 4, rue d’Enghien • 75010 Paris • Tél. (33) 01 53 24 14 50 • Fax (33) 01 53 24 14 51
E-mail : [email protected] • Site : www.dial.ird
Does Managerial Capital also Matter
Among Micro and Small Firms in
Developing Countries?
Axel Demenet1
10/2016
JEL: O17, M1
Abstract
The lack of managerial capital was recently put forward as a constraint
for developing countries firms (Bruhn et al., 2010). While established for
large and medium firms, its importance for micro enterprises has yet to
be proven: evidence found in Development Economics and
Entrepreneurial Studies is, at best, mixed. This paper uses a panel of
Vietnamese micro, small and medium enterprises to investigate this
question in a comparative manner. The data let building a multidimensional measure of Managerial Capital, and allows consistent
estimates of firm-level productivity. Even though bias might still affect
the estimation of the average influence of managerial capital on
productivity, I am able to show that this influence is as important for
micro firms as it is for medium ones.
Acknowledgments
Previous versions of this paper were meaningfully enriched by comments received at the 84th
ACFAS conference, Montreal; the STBI seminar at the UEH, and the 7th Vietnam Economist
Annual Meeting, Ho Chi Minh City.
1
[email protected]
(1) DIAL, UMR 225, IRD. 4 rue d’Enghien, 75010 Paris, France
(2) PSL Research University, Université Paris-Dauphine, LEDa, F-75016 Paris, France
1. Introduction
Neither husband nor wife knew how to read —a slight defect of education, which did not
prevent them from ciphering admirably and doing a most flourishing business. […] To relieve
himself of the necessity of keeping books and accounts, he bought and sold for cash only.
Balzac, Le Curé de Village, 1841
Few will question the relevance of advertisement expenditures for a multinational manufacturing
corporation. More eyebrows may rise upon asking the same question in the case of a single
informal sector worker producing rubber sandals. The motivation of this paper is to ask this
question, and to do so in a comparative manner: does managerial capital (MC) have the same
effect on productivity among micro, small and medium firms? While past and on-going research
proved the relevance of managerial capital for large or medium enterprises, the population of
micro and small enterprises was largely left aside.
Given the weight of this segment in total employment, and to the extent the long-awaited
development process is to occur through productivity gains, it is yet of prior interest to
understand the mechanisms that foster (or limit) their expansion. Several types of constraints
have already been put forward. Access to savings (Dupas & Robinson, 2013), access to finance
(de Mel et al., 2008), and human capital (Hsieh & Klenow, 2009) are among the more
documented ones. The lack of managerial capital, by contrast, only recently emerged as such
(Bruhn et al., 2010).
Micro, small and medium enterprises (MSME) in the developing world rarely use what is
considered elementary business practices in industrialised countries. The majority do not keep
basic written accounts, and they compete mostly with other local household businesses. Yet,
micro entrepreneurs themselves mention factors such as “keeping and interpreting financial
records” or “promoting product” as favouring business success (Bradford, 2007). A segment of
enterprises does display high organisational and managerial abilities. This heterogeneity in
managerial capital endowment can enter the production function as an additional efficiency
factor: it could be argued that even among micro-firms, managerial inputs could improve the
productivity of other inputs. Competing more aggressively, advertising products, incentivising the
wage determination or innovating could lead to higher value added.
Proving a causal relationship between productivity and managerial capital is challenging, as the
latter does not offer exogenous variations. It is rather part of the often-blamed (and ineluctably
2
unobserved) “ability” of the firm’s operator, and any link found with productivity can be
attributed to some other unobserved factor. The approach of this paper is accordingly not to
argue in the sense of a causal link, which observational data would struggle to back up. It is rather
to provide a comparative estimation of the impact of managerial capital on productivity by firm
size, using a synthetic indicator of MC and consistent productivity estimates.
To do so, I rely on a panel of Vietnamese MSME that let measuring several aspects of MC. I
propose a multidimensional measure of MC based on twelve variables, used to compute a
weighted score. I start by estimating the productivity of household firms, controlling for
simultaneity and input price bias. I then investigate the effect of MC on firm-level productivity,
controlling for unvarying heterogeneity, and testing for heterogeneity of the effect by firm size
categories. While larger firms are found to be on average more productive than smaller ones, I
find that the effect of MC is still significant –and of comparable magnitude– among the smallest.
Micro and small firms that have higher managerial abilities are indeed more efficient than others
–and they are as much more efficient as medium-size firms. I further investigate the separated
effect of the MC indicators and find the effect to be mainly driven by the firms’ ability to
compete aggressively.
The remainder of the paper reviews the literature (sect. I), presents the data (sect. II) and the
empirical measures of productivity and managerial capital (sect. III). Section IV presents the
estimation results of the link between MC and productivity, and section V presents robustness
tests.
3
2. Literature review: what do we know about micro and small enterprises’
managerial capital?
What exactly constitutes managerial capital? The notion has no widely accepted definition and
necessarily borrows from several fields of studies, which have complementary definitions. As put
by Syverson (2011) “Managers are conductors of an input orchestra”: defining managerial capital then
amounts to measuring the length of the conductor's baton; but it could also relate to the
conductor’s attitude and psychological traits. In the related literature that recently surged in the
field of Development Economics, MC is persistently proxied by business practices (and among
micro and small firms, by elementary business skills). The field of Management studies, in which
the focus has long been on the managers’ influence, also considers additional features of MC that
relate to the entrepreneurial spirit. Taken broadly, MC thus refers to all practices and traits of the
enterprise operator that potentially have an influence on the firms’ efficiency. As such, it can
include formal bookkeeping, inventory management, financial or strategic planning, and pricing
strategy; but also innovativeness, or self-confidence.
This section reviews the literature to determine which skills, practices or characteristics can best
proxy managerial capital in the case of micro and small firms, and to what extent their influence
on performance is established at this point. The Development Economics literature saw a recent
surge in business training programs’ evaluation, which provided insight into the relevance of
some business practices. The following review focuses on whether this literature did succeed in
proving the relevance of managerial capital for micro and small firms, rather than on the type of
programs that work. On the other hand, contributions are found in the field of Management and
entrepreneurial studies, where the relevance of the notion of Entrepreneurial Orientation (EO)
among micro firms in developing countries has received recent empirical attention.
A. Business practices in Development Economics literature: mixed evidence from
training programs
Interest for the role of the manager recently sparked in the field. A set of papers defined what is
considered “good practices” for large enterprises, and proved the effects of these practices on
productivity (Syverson, 2011; Bloom et al., 2010, 2013a, 2013b). In the case of household micro
and small firms, the picture is fuzzier.
4
Evidence mainly stems from programs’ evaluations. Numerous training programs have been
launched to improve microenterprises’ business skills around the developing world, and a
substantial impact evaluation literature jointly emerged. They cover a large and varying set of
skills. The more frequent ones include marketing, advertising, bookkeeping, elaborating growth
strategies, financial planning, costing, and suppliers’ optimisation. Most of these contributions
focus on two types of outcomes: in a first step, they ask whether training programs do improve
the skills that they teach, an in a second step whether these programs have, through improved
skills, an effect on firms outcome. The first step is a necessary condition to be able to attribute
any lasting improvement to the program itself. It is however not sufficient: improving practices
per se does not provide ground for support policies if no effect on outcomes are found.
Many of the recent papers in this strand yet fall into this category, showing only significant
improvements in business practices that do not translate into higher profits, faster growth or
improved survival (De mel, McKenzie & Woodruff, 2014, Fairlie et al., 2015; Fiala, 2014; Karlan
& Valvidia, 2011; Karlan, Knight & Udry, 2014). Business skills would then be teachable, but it is
harder to draw lessons on their intrinsic relevance. Not only the lack of significance, but also the
heterogeneity of interventions and contexts described in these studies is at stake. In addition to
varying in contents, the programs evaluated differ in terms of target populations, scale, duration,
means, and implementation schemes (some are combined with in-cash grants). The population of
firms heterogeneous: many studies consequently lack sufficient power to detect any effect on
outcomes. Among the studies having sufficient power to detect impacts, two papers found
impacts of trainings on short-term profits and sales: Berge, Bjorvatn, & Tungodden (2014) by
combining survey data and lab experiment, and De Mel, McKenzie, & Woodruff (2014) by
evaluating a combination of training and grants.
A review of this literature by McKenzie & Woodruff (2014), including 20 studies (among which
16 randomized control trials), logically concludes that “there is little evidence to help guide policymakers
as to whether any impacts found come from […] productivity improvements, and little evidence to guide the
development of the provision of training”. A review by Grimm & Paffhausen (2015) also point the lack
of power of most of the RCT included: 5 out of the 17 studies on all topics are able to detect
effect sizes of 20% and less. This might largely explain why “the training interventions do not generally
succeed in increasing profits”. Cho & Honorati (2014), using meta-regressions, draw a similar
conclusion. Programs aiming at improving microenterprises’ business skills in developing
countries succeed in doing so, especially among youth, but this literature fails to prove that they
translate into an improved growth or profitability. The intuitive assumption that the relevant
5
skills for micro and small enterprises are elementary compared to those of larger firms was
however evidenced (Drexler et al., 2014): when comparing two programs, only the effect of the
simplest (“rule of thumb”) is much larger among micro-entreprises.
Lessons drawn from a collection of programs evaluations mainly tell which skills can be
influenced by training programs. Conditionally on having sufficient power, they subsequently tell
whether these programs have an impact on microenterprises outcomes. But most of the
programs are small, involving typically 100 to 300 heterogeneous firms. As such, they may be
suited to improve the design and provision of future programs in related contexts, but are overall
insufficient. Drawing a more general lesson on the importance of Managerial Capital requires
going beyond the collection of trainings. Evidence from larger scale surveys could add value in
this context (McKenzie, 2011).
B. Entrepreneurial Orientation and micro enterprises performance
Together with business practices, manager’s traits and attitudes could matter. Another topic
covered by some business trainings that unsurprisingly found more popularity in Managerial
studies than in Economics is the analysis of attitudes and psychological factors. Not only
elementary business skills, but also the proactiveness and perseverance of business owners can
increase their performance (Glaub & Frese, 2011).
This finding echoes a second and complementary strand of literature on the link between
management and microenterprises performance. An equivalent of what is called in the
Economics literature “managerial ability”, which relates to attitudes, is the concept of
Entrepreneurial Orientation (Miller, 1983; Covin & Slevin, 1989). It can be measured at the firm
level
along
five
dimensions:
proactiveness,
innovativeness,
risk-taking,
competitive
aggressiveness, and autonomy of workers (Lumpkin & Dess, 1996). A recent set of papers aimed
at providing empirical evidence on the link between EO and microenterprises performance. In
the cases of Mexico, Argentina, Malaysia and the Philippines (respectively Campos et al., 2013;
Berrone et al., 2014; Lindsay et al., 2014; Munoz et al., 2015), the results suggest a positive
association between EO and performance. Besides converging results, these studies however
share numerous and strong methodological shortcomings. First, samples are small, ranging from
151 to 735 observations (for the latter with a 46% response rate) and generally non-random, with
on-site or administrative identification of respondents, which likely results in sampling errors.
6
Second, questionnaires are often self-administered or mailed. While this is not uniformly
associated with poorer data, the accuracy of the measure of micro and small firms’ performance
is more questionable. Let alone by the fact that mailing questionnaires requires a fixed address,
which many informal microenterprises lack. All of the cited papers rely on self-reported
appreciations of performance, using categorical variables, consistency of which is highly
questionable. 2 Further meta-analysis, as done by Rauch et al. (2009), is in turn problematic.
Among the 53 samples used, the average sample size is just below 270 respondents and only
seven papers use a dependent variable that is not a categorical “perceived performance”.
Findings from the evaluations of training programs aiming at improving business practices found
little significant improvements in performance. Similarly, the empirical investigations on the role
of entrepreneurial orientation (which approximates the part of managerial ability related to
attitudes) among microenterprises suffer from too many methodological shortcomings to
provide convincing evidence. Both aspects of MC are thus far discussed largely separately in the
two disciplines, probably because the assumption is that entrepreneurship matters little among
micro and small enterprises in developing countries. There is room for a closer look at the link
between managerial capital, understood as both practices and traits, and productivity. More
specifically, the question raised at this point is the potential heterogeneity of the link between
managerial capital and firm’s size. The only major contribution in this regard comes again from
McKenzie & Woodruff (2015), who look at the influence of business practices in marketing,
stock-keeping, record-keeping and financial planning on productivity. Their result, somehow
contradicting the lack of effects of business trainings, is that these practices explain as much
variation in outcomes in microenterprises as in large firms. The items included in the scorecard
on which they rely partly differ from the variables included in this paper (noticeably on
entrepreneurship variables). Together with the fact that Vietnam is not among the countries
covered, and the different approach I use (relying on panel data and productivity estimates),
makes the comparison of both papers useful.
2"In"general,"relying"on"self0reported"perceived"categorical"outcomes"is"problematic"for"comparison."In"
some"cases,"the"very"questions"used"for"self0reported"levels"of"outcomes"are"beyond"the"reporting"capacity"
of"many"microentrepreneurs,"e.g."when"asking"directly"to"report"the"“return"on"capital"employed”"or"
“growth"of"the"company’s"value”"0found"in"Campos"et+al.,"2013;"Munoz"et+al.,"2015."
7
3. Data
This paper relies on a rich panel data of MSME, including a majority of micro and small informal
firms, and several indicators of both business practices and entrepreneurial attitudes. This section
describes the survey data used (A) and the construction of the Managerial Capital index (B).
A. The SME panel
I use a panel of household firms, including 5,878 observations of 1,892 unique firms between
2007 and 2013. It is based on the manufacturing SMEs panel data collected in the frame of the
remarkably long-term partnership between the University of Copenhagen and Vietnamese
research institutes. The project involved the Central Institute for Economic Management (CIEM)
of the Ministry of Planning and Investment of Vietnam (MPI); the Institute of Labour Science
and Social Affairs (ILSSA) of the Ministry of Labour, Invalids and Social Affairs of Vietnam
(MoLISA); and the Development Economics Research Group (DERG) of the University of
Copenhagen. The survey has been initially funded by the Royal Embassy of Denmark in Vietnam
under the Business Sector Programme Support (BSPS).3 The panel data has been extensively used
in academic research (Rand & Tarp, 2011; Rand & Torm, 2012; Rand & Tarp, 2012).
The small and medium enterprise survey has been conducted nine times, more recently in 2005,
2007, 2009, 2011, 2013, and 2015. As the 2015 round is not available yet, and since the questions
of interest have changed after the 2005 round, I only use the 2007 to 2013 rounds. The resulting
dataset combines four rounds of survey, one every two years. For each round, outputs are
available for years n and n-1. The sample size for each round is initially around 2,500 firms, from
which only household firms were kept in the present sample. Observations present only one year
were dropped. The resulting sample includes 5, 878 observations.
The surveys were conducted in 10 provinces in Vietnam. The stratification was initially based on
the type of ownership, and aimed at reproducing the structure found in the General Statistics
Office (GSO) figures, and other pre-existing SME surveys in Vietnam such as Sakai and Takada
(2000). A significant share of informal firms was explicitly included, using on-site identification.
Further renewal of the sample implied selecting firms from both the new GSO listings and the
on-site identified informal firms. As a result, the SME surveys are not representative of the
informal sector in Vietnam as they are biased towards larger firms, but they rather provide a
3
More information can be found at https://www.wider.unu.edu/project/small-and-medium-enterprise-sme-survey
8
picture of the Micro, Small and Medium enterprises segment. Since the initial sampling in 2005
was based on a pre-existing survey from 1997, there is also “a slight bias against young, newly
established enterprises” (Rand et al., 2004).
A notable feature is the stability of the questionnaire over time. It consists of three modules: (i) a
main enterprise questionnaire; (ii) an ‘employee module’; and (iii) an ‘economic accounts module’.
Information collected includes firm’s and owner/managers’ characteristics, as well as detailed
records of production, sales and inputs. It also includes a rich set of indicators of Managerial
Capital indicators in the enterprise questionnaire.
Table 1 provides a description of the sample: number of observations per year, number of firms
in each size category, outcome and income levels. It also describes all the control variables that
are used in the empirical analysis. Descriptive statistics of managerial capital indicators are
provided below.
Table 1. Sample description.
2007
Observations
2009
1,439
2011
1,677
2013
1,538
1,224
mean
sd
mean
sd
mean
sd
mean
sd
Size: [1,2] workers (inc. owner)
0.22
0.41
0.26
0.44
0.30
0.46
0.30
0.46
Size ]2,5]
0.42
0.49
0.39
0.49
0.42
0.49
0.44
0.50
Size ]5,10]
0.23
0.42
0.22
0.41
0.18
0.38
0.19
0.39
Size 11+
0.13
0.34
0.13
0.33
0.10
0.30
0.07
0.26
Average size
6.54
7.67
6.42
7.69
5.55
6.22
5.00
4.72
Real value added
251
413
250
418
288
447
224
280
Real capital
1639
4160
1425
2498
2313
5823
1226
1869
Informal (BRC)
0.44
0.50
0.51
0.50
0.43
0.50
0.39
0.49
Premise: residential
0.37
0.48
0.35
0.48
0.33
0.47
0.21
0.41
Premise: production
0.38
0.49
0.44
0.50
0.45
0.50
0.49
0.50
Premise: only production
0.25
0.43
0.21
0.41
0.22
0.42
0.30
0.46
Road access
0.69
0.46
0.73
0.45
0.70
0.46
0.81
0.39
Manager: male
0.68
0.47
0.68
0.47
0.67
0.47
0.68
0.47
Higher secondary educ. or more
0.40
0.49
0.43
0.50
0.45
0.50
0.55
0.50
Food and beverages
0.35
0.48
0.35
0.48
0.38
0.49
0.38
0.49
Fabricated metal prod.
0.18
0.38
0.18
0.39
0.19
0.39
0.19
0.39
Wood
0.14
0.35
0.14
0.34
0.13
0.33
0.12
0.32
Furniture, jewellery, music equ.
0.08
0.28
0.07
0.26
0.08
0.28
0.09
0.29
Non-metallic mineral prod.
0.06
0.24
0.05
0.23
0.04
0.20
0.04
0.19
Textile
0.04
0.19
0.04
0.20
0.04
0.20
0.04
0.20
Apparel, leather
0.05
0.21
0.04
0.20
0.05
0.22
0.05
0.22
Other
0.10
0.31
0.11
0.32
0.09
0.29
0.09
0.29
Sector
9
The sample is primarily made of micro-firms. Household businesses of less than five workers
consistently represent between 63 and 74 per cent of observations, depending on the year. In
addition, 22 to 30 per cent of firms only include one or two workers, that is, own-account
workers or employers with a single additional employee. Medium-size household firms are also
included; they have up to a hundred employees, and provide a basis for comparison of larger size
and comparable status. Informal household firms, defined as having no business registration
certificate, represent 39 to 51 per cent of firms per year, which shows that the sampling
methodology (and renewal procedure) performed well in including those units, even though no
representativeness is claimed. The average size in the sample is 6.1 workers. Less than a quarter
of firms operate in premises that are dedicated to production only, which means that the majority
operates at home or in shared spaces. More than two thirds of the managers are males, and half
of them (averaged for all years) reached an education level above higher secondary. Firms
included in the sample are manufacturing, and the main sectors of operations are food and
beverages (with 37 per cent of firms), and metal products (19 per cent).
B. A multidimensional measure of managerial capital
Managerial capital, as discussed in section I., relates to all practices and traits of business
operators that have the potential to influence the firms’ efficiency. An advantage of the SME
survey in this regard is to provide a large range of indicators linked with business practices, as
well as entrepreneurial attitudes. I use in this study a multi-dimensional measure of MC, based on
twelve variables that are found (and consistent) in all survey rounds. These are (1) the use of
formal accounts, (2) the use of advertisement means, (3) the pricing method, (4) wage
determination mode, (5) hiring mechanisms, (6) the use of outside services, (7) the location and
(8) type of suppliers, (9) improvements attempts of products and services; (10) innovation plans
for the near future, and finally past innovation in terms of (11) products or (12) processes. The
incidence of each indicator by firm size (in terms of number of workers, including the
owner/manager) is provided in table 2. The description of these indicators by size is already
telling, and lets seeing that managerial capital is increasing in firm size.
First, the bookkeeping variable indicates whether the respondent does “maintain a formal accounting
book”. Only 9 per cent of firms across years do keep such formal accounts, and the proportion is
strongly increasing in firm size. Among the largest household firms (11 workers or more), the
percentage reaches 25 per cent. It should be stressed that this variable captures the existence of a
complete set of accounts; a negative answer may not indicate the total absence of books, as firms
10
may keep simple records through personal notes. The difference that is captured is rather
between keeping elementary books (or none), and using a complete accounting system. Second, a
binary indication of marketing efforts is based on a positive answer to the question “do you
advertise your products?”. It mixes as such all practices, from door-to-door information to radio
or TV spots. It is kept binary rather than categorical, as media coverage is barely used among
firms that do advertise. Almost inexistent among micro firms (where it reaches 1 to 3 per cent),
advertisement is less rare among medium firms (7 per cent), but remains scarcely used. Third, the
pricing method is a categorical variable indicating whether the firm fixes its prices according to
costs, competitors, or bargaining. There is no significant difference by size, and overall three
quarters of firms base their decision on costs –precise determination of which is however not
always possible among micro firms. Fourth, the wage determination variable indicates if wages
are determined following other sectors’ rates, following local competitors’ rates, by individual
negotiations, by the paying capacity of the firm, or by none of these methods (which is
necessarily the case for almost 15 per cent of firms that have no employee). Most of the medium
sized firms indicate using individual negotiations, while micro firms, even among those paying
wages, predominantly report no fixed method for determining wages. Fifth, the hiring channel
indicates which mean firms favour upon hiring new workers. Formal competitive channels
(labour service centres, labour exchanges, newspapers) are still marginal. The overwhelming
majority relies on recommendations by friends or relatives, or personal contacts. Sixth, the use of
outside services is constructed as a binary indication of whether the firm buys the following
services from external providers: market information; information services on technology; tax,
audit or accounting; legal advice; human resources advice; or vocational training. Barely any firm
among the micro and small ones do so. Even if the percentages increases with size, only 8 per
cent of the largest household firms use such services. Seventh, the distance to the main supplier
is indicated as whether firms buy all products from the same commune, or whether part comes
from within the province, from other provinces, or a significant part from direct import. The
smaller the firm, the more local its suppliers are. Even though some firms do import products, it
remains a marginal share in their total suppliers. Eighth, the supplier type indicates if a share
(more than 10 per cent) of suppliers are households or the State rather than other private firms –
which increases for the latter among the larger firms to reach 22 per cent. Finally, the incidence
of all indicators of product improvement, and (planned or past) innovation strongly increase in
size. While 46 and 16 per cent of the larger firms improve their products or innovate in terms of
processes respectively, the corresponding figures for micro firms are 15 and 3 per cent.
11
Table 2. Components of Managerial Capital: incidence by firm size
Pricing
Formal accounts
[1,2]
]2,5]
]5,10]
11+
Total
[1,2]
]2,5]
]5,10]
11+
Total
[1,2]
]2,5]
]5,10]
11+
Total
[1,2]
]2,5]
]5,10]
11+
Total
[1,2]
]2,5]
]5,10]
11+
Total
Advertisement
% costs
As compet.
Less th.
compet.
Bargain
0.01
0.07
0.17
0.25
0.09
0.01
0.74
0.10
0.01
0.15
0.03
0.73
0.09
0.01
0.17
0.03
0.78
0.07
0.01
0.14
0.07
0.72
0.10
0.00
0.17
0.03
0.74
0.09
0.01
0.16
Wage determination method
Local
Set by
Paying
None
Local SOE
Agric. rates Negociation
Other
enterpr.
auth.
capacity
0.85
0.02
0.00
0.00
0.00
0.08
0.03
0.01
0.21
0.12
0.01
0.01
0.02
0.43
0.19
0.01
0.02
0.17
0.02
0.00
0.02
0.53
0.22
0.02
0.00
0.18
0.02
0.01
0.02
0.55
0.17
0.05
0.32
0.11
0.01
0.00
0.01
0.37
0.15
0.02
Hiring channel
None
Newpaper Labour exchange Friends Authorities Personnal cont. Service centre Other
0.83
0.01
0.01
0.07
0.00
0.08
0.00
0.01
0.25
0.01
0.02
0.34
0.01
0.34
0.01
0.02
0.05
0.02
0.02
0.51
0.01
0.34
0.02
0.03
0.02
0.04
0.03
0.57
0.01
0.26
0.03
0.04
0.34
0.02
0.02
0.33
0.01
0.26
0.01
0.02
Suppliers
location
Suppliers
type
Outside
services Commune
Province
Country
Import
Private ents.
Households
State
0.01
0.28
0.63
0.09
0.00
0.52
0.38
0.11
0.03
0.16
0.65
0.19
0.00
0.46
0.38
0.16
0.04
0.09
0.64
0.28
0.00
0.36
0.42
0.22
0.08
0.08
0.59
0.33
0.00
0.40
0.38
0.22
0.03
0.17
0.63
0.20
0.00
0.45
0.39
0.16
Products
Innovation:
Planning innovations
Innovation: products
improvement
processes
0.15
0.14
0.02
0.03
0.29
0.25
0.02
0.07
0.41
0.37
0.03
0.10
0.47
0.44
0.03
0.16
0.30
0.27
0.02
0.08
The dimensions included in the operational definition of managerial capital thus relate to
business skills and entrepreneurship at the same time (in particular to its innovation and
proactiveness dimensions). The discriminating power of most of the MC variables is high, with
little firms reporting what assumingly constitutes “good” business practices or entrepreneurial
behaviours. I build from these variables a synthetic indicator of managerial capital using Multiple
Correspondence Analysis (MCA) for each year, which is later used to estimate the effect of MC
on productivity. All categorical variables are transformed into dummies, for which all modalities
12
are kept in the MCA. There is often no element to determine a priori which modality indicates a
higher managerial capital for categorical variables (such as supplier type). Even if there was, it
could be changing by firm size. MCA builds a binary indicator matrix that shows the incidence of
each axis by firm, which is used to obtain weights from the factorial axis. A firm’s MC score is
thus calculated by the weighted sum of its responses, and can be noted:
!"! =
!!" !!
!!!,..,!
where MCi is the i-th observation’s managerial capital score, Dij the response of unit i to
dimension j, and Wj the MCA weight for the first axis applied to category j. Descriptive statistics
on the normalised score by year and firm size are provided in table 4. Detailed results of the
MCA are provided in appendix table 1. While the contribution of the binary variables
(bookkeeping, advertisement, outside services utilisation, and innovation) to the MC index is
straightforward, it is not the case of categorical indicators. The MCA results are informative in
this regard, as they reveal the direction and intensity of the contribution of each modality to the
axis. It turns out that those firms able to find suppliers in other provinces (or other countries,
though this case concerns few firms), those for which State-Owned Enterprises (SOE) represent
a significant share of suppliers, and those which compete aggressively (pricing “lower than
competitors”) have higher MC levels.
The distribution of the MC normalised score, provided in table 3, can be interpreted in relative
levels. Units with a higher score display a mix of more frequent business practices (keeping
written accounts, using outside services, fixing wage through individual negotiations or relatively
to local competitors) and more entrepreneurial traits (advertising, innovating through products or
technologies, competing more aggressively).
Table 3. Managerial capital normalised score by year and firm size
2007
2009
2011
2013
[1,2]
mean
-0.863
sd
0.540
mean
-0.834
sd
0.602
mean
-0.682
sd
0.626
mean
-0.846
sd
0.495
]2,5]
-0.049
0.801
-0.012
0.854
0.086
0.801
0.197
0.867
]5,10]
0.454
0.918
0.548
0.882
0.582
1.106
0.534
0.859
11+
0.764
1.184
0.761
1.007
0.617
1.220
0.912
1.238
All
0.000
1.000
0.000
1.000
0.000
1.000
0.000
1.000
0.021
1.020
0.043
1.014
0.017
1.006
d(MC)
13
The smaller the firm, the lower is the value of the managerial capital index. Micro firms
concentrate in the lowest score values, and few reach high levels of MC score. Medium
enterprises, which on average 21 workers, rank the highest. The increase is almost linear in size.
An additional question is the extent to which managerial capital varies within firms. This is all the
more important than part of the empirical strategy uses fixed effects regressions, which use
variations in the index for identification. This variation comes from two sources. On the one
hand, some firms changed owners between survey rounds. These changes may imply sudden
changes in managerial practices. On the other hand, firms may implement changes in practices
and/or entrepreneurial attitudes without changing owners. The within-firms variations in MC are
reported at the end of table 3, by the yearly average change and standard deviation.
14
4. Empirical strategy
The aim of this paper is to investigate the effect of managerial capital on productivity among
household firms. More importantly, it is to test the heterogeneity of this effect by firm size to
determine whether the concept does matter among the smallest firms. The overall strategy is thus
to evaluate the effect of MC on productivity, and then to test for heterogeneity.
If managerial capital does have an influence on productivity, it should allow firms to reach a
similar output level with fewer inputs – or conversely to increase output while inputs are kept
constant. In order to evaluate this link and to further test its relevance for the population of
micro firms, this paper relies on a two-stage estimation procedure. This approach amounts to I
investigating the influence of MC on Total Factor Productivity (TFP). The first stage consists in
estimating unbiased firm-level productivity. The second stage consists in regressing the MC index
on these estimated productivity levels, together with firm size indicators. Doing so, I obtain the
effect of MC on productivity net of firm size. The heterogeneity of its effect by size is further
tested by interaction terms. Comparing the importance of MC by firm size then amounts to
testing for heterogeneous effects in this regard. The parameter of interest is thus not only the
effect of MC in itself, but also its variation with firm size.
An alternative would be to estimate the direct influence of managerial capital in a production
function. But this would assume substitutability between MC and other inputs, which may be
true to a certain extent if efficiency gains are involved, but does not hold for low levels of capital
and labour. In a micro-firm with two workers, assuming that a higher MC can compensate for
one of them is probably too strong. The preferred approach is to assume that the influence of
MC is indirect, and goes through its interaction with other inputs.
Another limit of estimating direct production functions is the difficulty to argue in favour of a
causal influence. Even when using panel estimators, MC does not vary exogenously. Firms with
higher managerial capital may have higher levels of productivity for reasons that are not
observed, besides the effect of the former on the latter. The two-stage procedure, by contrast,
lets estimating unbiased TFP levels. Regressing MC on the firm-level productivity then requires
the lesser assumption that the bias plaguing the estimation of the effect of MC does not vary by
firm size. This indirect approach relies heavily on the first-step estimations of productivity, which
has to overcome a number of endogeneity concerns, and is described hereafter.
15
Estimating firm-level productivity
The empirical study of the link between MC and firm-level productivity can only be as good as
the first-stage estimations of this productivity. The correct identification of the production
functions, is among the oldest challenge in the empirical economic literature, and still evolves
rapidly. Recent complementary methods allow significant progress towards overcoming the
endogeneity challenges. Essentially, true productivity levels remain unobserved -and so are
productivity shocks, to which firms can react differently. As long as input levels are chosen in
relation with these unobserved determinants, OLS estimations will be biased. Specifying a CobbDouglas value added function:
!"!" = !! + !! !!" + !! !!" + !!" + ε!"
(1)
Where !"!" is the value added of firm i at time t. l and k, are respectively the labour and capital
inputs. All variables are transformed in natural logarithm. The error term has two components,
!!" and !!" , the former being correlated with inputs. The size and direction of the bias of OLS
coefficient on capital will depend on the correlation between inputs and productivity shocks, and
more crucially of whether this correlation varies with time. If it does not, the inclusion of firmlevel fixed-effects will yield consistent coefficients. Provided firms’ exit is also determined by this
unobserved but unvarying productivity, fixed effects (FE) will also solve potential selection bias
due to endogenous exits. The unvarying nature of the unobserved productivity could however be
a rather strong hypothesis, especially when using long-term panels. Other approaches allow for
inputs to be endogenous with respect to a time varying unobservable. Three contributions largely
frame the empirical literature in this regard: Olley & Pakes (1996), Levinsohn & Petrin4 (2003),
and Ackerberg, Caves & Frazer (2015) (henceforth OP, LP, and ACF respectively). The former
two rely on the relationship between some intermediate input entering the production function
and the unobserved productivity:
!!" = !! (!!" , !!" )
(2)
This function can be inverted, assuming in particular a monotonic increase in !!" so that the
productivity is a function of two observed inputs:
!!" = !! (!!" , !!" )
4
(3)
The LP method is usually revenue-based, but fits value added models as well.
16
Ollin and Park build on the idea that firms change investment (conditional on capital stock) in
response to productivity shocks, and provide a non-parametric representation of this inverse
function to estimate production functions in two stages. Investment might however not react
strongly to productivity shocks –or at least, not monotically- and the adjustments could take
place at other levels. Levinsohn & Petrin suggest using instead a more varying intermediate input
demand function such as material expenditure or energy costs. Ackerberg, Caves & Frazer (2015)
highlight a functional dependence problem of both OP and LP specifications, whereby labour
can be a deterministic function of the variables on which the first stage is conditioned. They
propose an alternative (“though quite related”) estimation strategy, where inverted input demand
functions are conditional on the choice of labour input. Wooldridge (2009) additionally proposes
a stacked version of LP’s moments, estimated by GMM with efficiency gains, again based on
unconditional input demands.5 Besides being more efficient than the two-step estimators, this
procedure can moreover correct for serial correlation (Van Beveren, 2010).
All these estimation techniques rely on structural assumptions that, despite progressively gaining
in generality, still largely condition their validity or failure. In practice, the specificities of the
firms’ population and of the available information therefore weights equally -or more- than the
overall performance of each technique in choosing the empirical approach. In the relatively
specific case of Vietnamese manufacturing SMEs, I use a combination of FE, LP and
Wooldridge’s (2009) estimations of production function in addition to the benchmark OLS
regressions.
OLS estimation of the firm-level value added function hence follows equation (1), including a
time trend. Assuming that the unobserved productivity is mostly fixed in time, I estimate the
same equation with firm-level fixed effects:
!"!" = !! + !! !!" + !! !!" + !! + !!"
(4)
From both regressions, we can predict the productivity levels by taking:
! = !(!!" − !! !!" − !! !!" )
(5)
Further controlling for the simultaneity of inputs using LP, OP or ACF requires additional
hypothesis on the evolution of productivity and on the timing of firms’ choices. As investment
only concerns 45.5% of firms across years, down to 33.7% among firms with 1 or 2 workers, it
5
Comprehensive reviews of production function estimation techniques include Eberhardt and Helmers (2010), or
Van Beveren (2010).
17
cannot be used as proxy without introducing selection bias. Among the available non-parametric
corrections, the LP procedure thus makes more sense, and electricity costs are used as
intermediate input proxy in the core of the paper. Productivity is then typically assumed to follow
a first-order Markov process: !!" = ! !!" !!"!! + !!" where !!" is uncorrelated with !!" but
can depend from !!" . The LP procedure then assumes that given !!" the firm will decide on !!" ,
and then determine !!" accordingly. The rearrangement of eq.1 is thus:
!"!" = !! !!" + !!" (!!" , !!" ) + !!"
(6)
Where:
!!" !!" , !!" = !! + !! !!" + !! (!!" , !!" ) (7)
and
! !!" !!!" , !!" , !!" = 0
(8)
The ACF critique essentially states that !! and !! are instead chosen simultaneously, which
plagues the identification of !! in the first stage. Following Wooldgridge (2009), the last and
preferred specification of productivity estimation in this paper consists in estimating !! and !!
directly by GMM. Assuming:
! !!" !!!" , !!" , !!" , !!"!! , !!"!! , !!"!! , … , !!! , !!! , !!! = 0 (9)
and restricting the dynamics of productivity shocks:
! !!" !!!" , !!"!! , !!"!! , !!"!! , … = ! !!" !!!"!! = ! !!"!! = !(! !!"!! , !!"!! ) (10)
We can write !!" = ! !!"!! + !!" with ! !!" !!!" , !!"!! , !!"!! , !!"!! , … = 0. In other words,
inputs !!" and !!" are correlated with productivity innovations !!" ; whereas !!" , which is set at
the previous period, is not. Neither are all past values of !!" , !!" and !!" . They provide a set of
instruments to identify !! and !! .
18
I estimate the four models (OLS, FE, LP, and Wooldridge) of value added function from the
unbalanced panel of firms.6 I use the log of deflated value added as outcome, and the log of total
employment at year n, and the log of real capital value at the end of the previous year as inputs.
LP model further includes the log of real electricity expenditures from the last period as proxy.
Wooldridge’s estimations of productivity rely on lagged values of !! and exponential functions of
log inputs and intermediate input. Concerns about the input prices bias, the endogenous exit of
firms and industry-specific effects may remain. Indeed, if firms face different input demand
functions (and/or operate at different points of the curves), the correction introduced by the
proxy intermediate input variable will further bias the results. Electricity costs are however
arguably similar across firms, and should not introduce further differences. Material expenditures
are used as robustness checks to estimate alternative productivity measures. Second, to the extent
less productive firms are more likely to exit the sample, it is still possible that the productivity
estimations will suffer from endogenous exit. The only method that directly corrects for this is
the OP estimation, and in practice, the corrections for firms exit are very small. 7 Lastly, a
common practice (challenged, among others, by Bernard, Redding and Schott; 2009) consists in
estimating separated production function by industry. Given the high concentration of our
sample firms within a few sectors (cf. table 1), industry-specific estimates would require grouping
arbitrarily sectors where few observations exist, and would result in introducing additional noise
rather than separating heterogeneous manufacturing firms. The productivity is thus estimated on
the full sample of firms.
Table 4 presents the coefficient estimates of firm-level production function on the whole
population of firms, using the unbalanced panel. Column 1 correspond to OLS estimation of
eq.1 with an additional time trend, column 2 further introduced fixed effects to estimate equation
(1); column 3 applies the LP procedure with electricity costs as intermediate input, and column 4
implements Wooldridge’s GMM estimation. Past values of inputs are limited to one lag in order
to prevent loosing years of observations. Past values for 2007 were obtained using the 2005
dataset. LP and Wooldrige results use fewer observations due to missing information on
intermediate input variables. Looking at excluded firms did not reveal any specific pattern. In
particular, excluded firms were not concentrated among the smallest microenterprises for which
one could fear that no electricity at all is used.
6
On implementing PF estimators in Stata, see Yasar, Raciborski & Poi (2008), Petrin, Poi & Levinsohn (2004).
7
Such is the case for Newman et al. (2015) using Vietnamese data on larger firms.
19
Table 4. Production function estimations
OLS
FE
LP
W
Log(capital)
0.242***
0.133***
0.148***
0.149***
(0.009)
(0.011)
(0.012)
(0.0106)
Log(#workers)
0.936***
0.711***
0.841***
0.913***
(0.018)
(0.025)
(0.017)
(0.0272)
time
0.064***
(0.008)
Constant
1.025***
2.272***
-0.738
(0.050)
(0.086)
(0.975)
Observations
5,699
5,699
4,925
4,825
P(constant returns to scale)
0.00
0.00
0.5518
0.01
*** p<0.01, ** p<0.05, * p<0.1
Clustered s.e. by firms in models 1,2,4, bootstrapped with 250 replications for LP estimates
Compared to the benchmark –and expectedly biased- OLS estimations, all corrections
consistently reduce the estimated returns to capital, which is consistent with the simultaneity bias
(more productive firms choosing more capital). The reduction in the labour coefficient is less
marked; heterogeneity in productivity levels could arguably allow employing more workers
altogether, or producing the same output with less workers. Fixed effects estimations overcorrect
the bias for both inputs, and find lower returns (and overall decreasing returns to scale).
However, the magnitude of firms’ productivity shock does not vary much with time in this
particular population of firms: overall, FE estimates are comparable with other corrected returns.
Both models correcting for fixed and varying simultaneity find returns to capital of nearly 15 per
cent at the mean, and the preferred Wooldridge specification yields high average returns to
labour, close to the OLS estimates. One explanation could lie in the proportion of micro-firms in
the sample for which more productive firms are those who can employ an additional worker. A
useful benchmark of productivity estimation on Vietnamese firms is found in Newman et al.
(2015), who use a sample of much larger firms and find returns of comparable range, varying by
industry. The differences in coefficients between estimation methods are comparable in sign and
size.
20
5. Results: managerial capital and productivity among micro-firms
The primary objective of this paper is to investigate the link between managerial capital and
productivity, and to test whether this link depends on firm size. Even if the firm-level
productivity levels are efficiently estimated, OLS regressions of MC on productivity are likely to
be biased by the exclusion of other factors entering total factor productivity. A key aspect is the
extensible boundaries of the concept of managerial capital. As MC essentially aims at measuring
owners’ abilities, it will be difficult to include a sufficient number of dimensions to arguably
capture all of it. Even when using the twelve variables available in the SME survey, it would be
difficult to argue in the sense of a perfect measure of MC. The strategy used in this paper is to get
as close to causal inference as possible using the whole population of firms, using panel
estimators to regress MC on productivity, and then to test whether the association is similar
across firm size. The assumption is that even though some varying heterogeneity could still
plague the estimation of the MC-productivity link, these potential biases can be constant by firm
size, and comparing the significance and size of the effects is hence feasible.
In order to control for all fixed determinants of both MC and productivity, whether observed or
not, all estimates use firm-level fixed effects. The outcome variable is the total factor productivity
estimated in section III.1. Specifically, I use the standardised productivity estimates of the
preferred specification (Wooldridge, 2009) in this section, and alternative productivity estimates
as robustness tests. The baseline specification is thus:
Ω!" = !! + !! !"!" + !! !!" + !" + !! + !!"
(11)
Where Ω!" represents firm-level productivity estimated by Wooldridge’s (2009) GMM, !"!" is
the managerial capital score of firm i at time t, !!" is the firm size category, and ! a vector of
controls for trends. Firms fixed effects are denoted !! . This set ! of control variables aims at
removing potentially differentiated evolutions by years, sectors, or regions (and finally with
time*regions interactions). As neither the productivity levels nor the MC scores can be directly
interpreted in levels, all results are provided as standardised variations. The results of this baseline
estimation are provided in the first column of table 5. I find that a standard deviation in MC
score results in a 6 per cent increase in productivity on average among the MSME sample,
significant at 1 per cent. This average effect is net of the influence of firm size.
A further concern regarding bias arises from the limited indicators of managerial capital available
in the data. As other practices and traits that could be used as additional proxies are not available
21
in our data, part of the manager’s ability remains unobserved by our MC score. I introduce a set
of controls to proxy individual ability, which shows some variation across year (firm owner does
change in some cases). Gender, education (higher secondary or more), and age class are
controlled for in the following model (included in the X vector). An additional set of timevarying firm characteristics is included: informality (being registered or not), type of premises,
and access to infrastructures (road, in this case). All can vary during the time period considered,
and may influence productivity.
Ω!" = !! + !! !"!" + !! !!" + !" + !" + !! + !!" (12)
Results of equation 12 are provided in column 2; additional controls have little influence on the
MC coefficient, which tends to indicate that fixed effects estimates removed most of the existing
bias. Looking at the description of MC score by size, we know that larger firms have both better
business practices, and more entrepreneurial attitudes. We also know from the results of section
III that productivity levels are strongly and constantly increasing with firm size. The results
provided in table 5 are net of this firm size effect, but the coefficient of MC score can
nevertheless have different slopes depending on the size category if its influence on
microenterprises is lower than in medium ones. A third model includes interactions between MC
index and size categories:
Ω!" = β! + !! !"!" + !! !!" + !" + !" + !(!"!" ∗ !!" ) + !! + !!"
(13)
The vector ! of estimated coefficients by size is then used to test for a differentiated effect of
MC depending on the category. Below table 5, I report Wald test for joint significance of the
interaction term.
22
Table 5. Managerial capital and productivity
Standardized values of (MC)
Size ]2,5]
Size ]5,10]
Size 11+
(1)
(2)
(3)
0.059***
(0.009)
0.559***
(0.019)
1.080***
(0.025)
1.725***
(0.039)
0.058***
(0.009)
0.576***
(0.018)
1.113***
(0.025)
1.759***
(0.039)
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.093***
(0.018)
0.547***
(0.022)
1.091***
(0.028)
1.731***
(0.050)
-0.040**
(0.019)
-0.051**
(0.025)
-0.042
(0.038)
Yes
Yes
Yes
Yes
Yes
-0.661***
(0.015)
-0.918***
(0.105)
-0.893***
(0.104)
4,840
0.603
1,850
4,840
0.604
1,850
(Size_cat=]2,5])*MC
(Size_cat=]5,10])*MC
(Size_cat=11+)*MC
Controls (ability)
Time dummies
Sector dummies
Region dummies
Time*region interaction
Constant
Observations
R-squared
Number of id
4,840
0.561
1,850
Testparm interactions: F(3, 1850) =1.79
Prob > F = 0.1467
As much as managerial capital and productivity depend on firm size, the effect of the former on
the latter is as important among micro firms as among medium ones. Marginal effects depending
on the size category and significance levels are reported in appendix 2. They confirm this finding.
The indicators of managerial capital, although more scarcely found among micro firms,
discriminate equally or more in this population between productive firms and subsistence
businesses. The effect of business practices and entrepreneurial orientation on productivity is
thus as large as among larger enterprises. In other words, these results show that when every
other factor (except some possibly remaining time varying heterogeneity) is controlled for,
including levels of inputs, firms with a higher MC generate more output. More importantly, they
show that the difference induced by a higher managerial capital is as large in micro-firms as in
small and medium ones.
23
6. Robustness
This section provides further robustness tests. Section A uses alternative measures of
productivity, and reproduces both steps of the two-stage procedure. The results are very
consistent across estimation methods. Section B applies the results of the alternative, direct
estimation where MC directly enters the production function. Although this is not the preferred
specification, the results are highly consistent. Finally, section C provides the results of a
comparable approach on a different dataset. Even while the SME panel includes a large number
of informal microenterprises, it is biased towards larger firms. An alternative dataset is thus
employed, which has the advantage of being representative of the Household Businesses
population (in the two cities where it was conducted: Hanoi and Ho Chi Minh city). The
drawback is that a much smaller number of variables are included in the MC index, and that the
data does not let estimating production functions as the capital measure is inconsistent.
A. Alternative productivity estimation
Most of the previous analysis relies on the firm-level productivity estimated in the first stage. A
remaining bias in these estimates would cast doubt on the results of the second stage, in
particular if the bias is somehow also correlated with managerial capital. The results were reobtained using an alternative estimation of productivity, based on a different correction for
simultaneity using material expenditures. The results of the production function estimations are
provided in table 6, using the same specifications than section 4.
Table 6. Production functions estimates using log material expenditures
OLS
FE
Log(capital)
Log(#workers)
time
Constant
Observations
0.242***
(0.009)
0.936***
(0.018)
0.064***
(0.008)
1.745***
(0.055)
5,686
0.133***
(0.011)
0.711***
(0.025)
2.997***
(0.078)
5,686
LP
W
0.158***
(0.012)
0.835***
(0.016)
0.171***
(0.0105)
0.872***
(0.0256)
4,885
1.303***
(0.416)
4,875
Returns to capital and labour go through overall comparable corrections when using this
alternative proxy of material expenditures, although the coefficients of capital of LP and
Wooldridge estimators are higher (and the returns to labour are lower). The firm-level
productivity is then used as outcome variable and the results of similar regressions than those of
24
section IV are provided in table 7. The effect of MC is consistent and still yields a 4% increase in
productivity per standard deviation. The heterogeneity of the coefficient by firm size is, again,
impossible to back up. Joint significance test of the interaction terms (reported below table 8) do
not let concluding to their significant influence. Managerial capital does matter among small
firms, at least as much as it does for medium-size ones. Finally, productivity estimates using other
corrections than the preferred Wooldridge specification (fixed effects and LP) were used with
similar results.
Table 7. Managerial capital score and alternative productivity measure.
Standardized values of (MC)
Size ]2,5]
Size ]5,10]
Size 11+
(1)
0.069***
(0.010)
0.514***
(0.020)
1.006***
(0.028)
1.622***
(0.041)
(2)
0.067***
(0.010)
0.539***
(0.019)
1.049***
(0.027)
1.660***
(0.041)
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
(3)
0.097***
(0.020)
0.514***
(0.023)
1.035***
(0.032)
1.642***
(0.051)
-0.031
(0.019)
-0.051*
(0.028)
-0.042
(0.039)
Yes
Yes
Yes
Yes
Yes
-0.615***
(0.017)
-0.926***
(0.124)
-0.903***
(0.123)
4,889
0.559
1,852
4,889
0.559
1,852
(Size_cat=2)*MC
(Size_cat=3)*MC
(Size_cat=4)*MC
Controls
Time dummies
Sector dummies
Region dummies
Time*region interaction
Constant
Observations
R-squared
Number of id
4,889
0.483
1,852
Testparm interactions: F(3, 1849) = 0.78
Prob > F = 0.5061
25
B. Direct estimation
Notwithstanding substitutability considerations, it can be of interest to compare the results of
direct estimations where MC enters the production function as an additional input. In line with
the previous approach, the actual question of interest would then be the significance of its
interaction term with inputs. Joint significance would imply that its effect is heterogeneous by
input levels. Taking the empirical counterpart of a Cobb-Douglas production function, we can
write:
VA!" = !! + !! !"!" + !! !!" + !! (!"!" ∗ !!" ) + !! !!"
+ !! !"!" ∗ !!" +!" + !" + !! + !!"
Using the same notations as above, and similarly taking natural logarithms of value added and
inputs. MC index and outputs are again expressed in standardised values. The direct influence of
managerial capital on output is !! . Its effects might vary with firm size at different levels of
capital and labour, in which case !! or !! will be significant. This would imply that the direct
influence of MC on output is not linear along levels of capital and/or number of workers. Table
8 provides the results of various declinations of equation 14. The first column uses no interaction
terms, and no controls. The second column introduces time-varying controls, with little impact.
Variations in managerial capital explain, on average, 9 to 11.5 standard deviation in output.
Interaction terms are, again, jointly unsignificant.
26
Table 8. Direct effect of Managerial capital and heterogeneity by input levels
(1)
MC
(2)
(3)
0.090***
(0.012)
0.674***
(0.026)
0.135***
(0.011)
0.092***
(0.012)
0.681***
(0.028)
0.104***
(0.011)
Controls
Time dummies
Sector dummies
Region dummies
Time*region interaction
Constant
No
Yes
Yes
Yes
Yes
3.042***
(0.076)
Yes
Yes
Yes
Yes
Yes
3.328***
(0.168)
0.115**
(0.057)
0.696***
(0.030)
0.102***
(0.011)
0.081**
(0.041)
0.055
(0.045)
0.072
(0.048)
-0.012
(0.009)
Yes
Yes
Yes
Yes
Yes
3.312***
(0.169)
Observations
R-squared
Number of id
5,699
0.283
1,889
5,699
0.342
1,889
5,699
0.344
1,889
log(# workers)
log(capital)
MC * (Size ]2,5] )
MC * (Size ]5,10] )
MC * (Size 11+ )
MC* log(capital)
Testparm interactions: F(3, 1849) = 1.42
Prob > F = 0.2337
27
C. Alternative sample: the Household Business and Informal Sector survey
Finally, one could be concerned by the fact that the data is biased towards larger firms. The
inclusion of informal micro enterprises is non-random, and the sample is overall not
representative. I conducted a similar analysis on an alternative panel dataset, the Household
Business and Informal Sector Survey (HB&ISS). These data are representative of the Vietnamese
informal sector, with an average firm size of 1.8 workers (including the owner). They were
collected in 2007 and 2009 in Hanoi and Ho Chi Minh City. They contain closely related
indicators of managerial capital, indicating: (1) if the firm adopted a strategy, (2) if the owner
prospects actively customers (by advertising or “trying to make him/herself known” among
neighbours), (3) if he/she is able to react to shocks (4) if he/she sets prices by bargaining or
related to production costs (compared with being price-taker), and (5) is he/she keeps no
accounts, simple records, or formal books (which answers some concerns about the bookkeeping
variable used in the previous analysis). Table 9 presents the incidence of these alternative
indicators in the HB&ISS data, and the MC index constructed with a similar methodology. A
comparable proportion of micro firms use formal accounts (3.3 per cent) –but 40 per cent use
personal notes rather than no accounts at all. The other indicators of MC are defined more
loosely than above, allowing for instance “bargain with customers” to be considered “active
pricing”, or defining “prospecting customers” as those doing anything but “waiting for them to
show up”. This explains their higher incidence in the population. The relative values of the MC
score, measuring distances between levels of managerial capital, yet bear the same meaning.
Table 9. MC index components in the HB&ISS sample
min
max
0.000
Standard
deviation
1.000
-1.366
2.914
Has a strategy
0.242
0.428
0
1
Prospect customers
0.418
0.493
0
1
Shock reaction
0.285
0.452
0
1
Active pricing
0.755
0.43
0
1
No accounts
0.571
0.495
0
1
Personal notes
0.396
0.489
0
1
Formal accounts
0.033
0.18
0
1
Mean
MC index
Accounts :
The panel being limited to two years of survey, together with inconsistencies in the measure of
capital, do not let estimating similar production functions. Instead, I estimate direct production
functions models with firm-level fixed effects, using the log of net profits as outcome variable,
and controlling for a large number of varying characteristics. Results using the balanced panel of
28
2,008 household businesses are provided in appendix table 3. The set of controls progressively
included in columns 1 and 2 includes informality, size, type of premises, and access to basic
equipment such as water or electricity. Among these household businesses, representative of the
Vietnamese urban informal sector, the MC score also has a positive and statistically significant
association with performance. This association ranges from 8 to 9 per cent.
29
7. Conclusion and discussion
This paper provides a straightforward answer to an open question of growing importance. It uses
a rich panel data consisting of 5,878 observations of 1,892 unique firms surveyed between 2007
and 2013, in which a set of indicators of managerial capital is available -and consistent across
years. One original feature is that these indicators allow combining standard indicators of
business practices and less frequent indicators of the firm’s entrepreneurial orientation into a
single score of MC. The results of consistent firm-level productivity estimates are regressed on
this score. A variation of one standard deviation in the MC score is associated with a 9 to 11 per
cent significant increase in productivity. The interaction terms with firm size category are not
significant, as confirmed by further marginal results. Again, no causal statement is made on the
relation between our measure of MC and productivity. Even though a large number of biases are
technically controlled for, variations in managerial capital remain unexplained. Rather, the
statement is that managerial capital does matter among micro and small firms, at least as much as
it matters among medium ones.
The implications of these results are, by contrast, not straightforward. Using observational data
allows using a more complex measure of MC, including elements that one cannot exogenously
change with a training program. On the other hand, it does not provide directly usable keys to
enhance micro-firm productivity. A key preliminary question would be to determine if managerial
capital, and in particular its entrepreneurial orientation component, is teachable at all. If the
observed variations find no explanation, one would be left with the frustrating justification of
unobserved individual talent (as Balzac attributes to monsieur Graslin, whose education level takes
no part in explaining entrepreneurial talent). A rough indication is given in table 10, which
regresses MC score on the set of available individual characteristics with OLS to provide some
insights on its variation.
Less than 20 per cent of the variance of the MC score is explained by the (limited number of)
individual characteristics. Yet, besides males and younger individuals having higher MC scores,
education has by far the largest influence. The little variance explained can be interpreted as a
proof of the relevance of the MC index – which indicates more than differences in education.
30
Table 10. Explaining managerial capital
Male
Finished lower secondary
Finished upper secondary
Age 30-45
Age 46-60
Age >60
HH size 3-5
HH size>6
Constant
Observations
R-squared
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
0.222***
(0.034)
0.102**
(0.042)
0.424***
(0.044)
0.011
(0.038)
-0.017
(0.046)
-0.127*
(0.072)
-0.281***
(0.073)
-0.362***
(0.084)
0.149
(0.104)
5,878
0.197
A useful comparison to challenge the validity of these results is the paper of McKenzie and
Woodruff (2015), which evaluates the link between business practices and small firms survival
and profits. Besides using multiple countries, one advantage of their paper is to include a larger
set of business practices. On the other hand, they do not estimate unbiased firm-level
productivity, and they focus only on business practices, including no indication of entrepreneurial
orientation or attitudes. Their message is overall comparable: managerial capital also matters for
small firms in developing countries. Furthermore, the variation in MC score is strictly comparable
among their size groups, as when larger firms (20 workers or more) have a one-standard
deviation higher score. The main difference lies in the size of the effects. McKenzie and
Woodruff’s (2015) effect of MC on profits are large (22 per cent at the mean for a one sd. in
MC), which, they argue, echoes the findings of Bloom and Van Reenen (2007). The effect found
in this paper, although significant, is half of that (9 to 11 per cent). A seducing explanation lies in
the sharper productivity measures used in the present paper. The magnitude of the effects found
is easier to reconcile with the mixed results found in training programs evaluation, many of which
lack power to detect small effects. It is nevertheless likely that the explanation lies in the different
countries included (Vietnam is not part of their study), and in the additional indicators included
in the MC index. Their paper uses a more detailed set of variables related to market research
31
capacity, stock control, and financial planning. This additional information probably captures, at
least partly, more variation in the practices that influence small business productivity.
Determining which skill or trait should be promoted is undoubtedly important from a policy
perspective, and the accumulation of training programs in heterogeneous contexts should, at one
point, come up with consensual findings. The approach of this paper is more general in the sense
that it rather evidences that managerial capital is a relevant prism to look at the efficiency of
micro and small firms in developing countries.
32
References
Ackerberg, D.A., Caves, K, & Frazer, G. (2015). Identification Properties of Recent Production Function
Estimators. Econometrica, 83(6), 2411–2451.
Berge, L.I.O., Bjorvatn, K., & Tungodden, B. (2011). Human and financial capital for microenterprise
development: Evidence from a field and lab experiment. Management Science, 61(4), 707 – 722.
Bernard, A. B., Redding, S., & Schott, P.K. (2009). Products and productivity. Scandinavian Journal of
Economics, 111(4), 681–709.
Berrone, P., Gertel, H., Giuliodori, R., Bernard, L., & Meiners, E. (2014). Determinants of Performance in
Microenterprises: Preliminary Evidence from Argentina. Journal of Small Business Management. 52(3): 477–
500.
Bloom, N., Eifert, B., Mahajan, A., McKenzie, D., & Roberts, J. (2013a). Does Management Matter?
Evidence from India. The Quarterly Journal of Economics 128 (1): 1-51.
Bloom, N., Schankerman, M., Van Reenen, J. (2013b). Identifying technology spillovers and product
market rivalry. Econometrica Volume 81, Issue 4, pages 1347–1393.
Bloom, N., Mahajan, A., McKenzie, D., Roberts, J. (2010). Why do firms in developing countries have low
productivity? American Economic Review: Papers & Proceedings, 100(2), 619–623.
Bloom, N., & van Reenen, J. (2007). Measuring and explaining management practices across firms and
countries. Quarterly Journal of Economics, 122(4), 1351–1408.
Bradford, W.D. (2007). Distinguishing Economically from Legally Formal Firms: Targeting Business
Support to Entrepreneurs in South Africa’s Townships. Journal of Small Business Management, 45(1): 94–115.
Bruhn, M., Karlan, D., & Schoar, A. (2010). What capital is missing in developing countries? American
Economic Review: Papers & Proceedings, 100(2), 629–633.
Campos, H.M., Acuña, L.S.A., de la Parra, J.P.N., & Valenzuela, F.A.A.( 2013). Entrepreneurial
Orientation in Mexican Microenterprises. Journal of Entrepreneurship, Management and Innovation. 9(3): 5-20.
Cho, Y., & Honorati, M. (2014). Entrepreneurship programs in developing countries: A meta regression
analysis. Labour Economics,28, 110–130.
Covin, J. G., & Slevin, D. P. (1989). Strategic management of small firms in hostile and benign
environments. Strategic Management Journal. 10: 75–87.
de Mel S., McKenzie D., & Woodruff C. (2014). Business Training and Female Enterprise Start-up,
Growth, and Dynamics: Experimental evidence from Sri Lanka. Journal of Development Economics, (106): 199210.
de Mel, S., McKenzie, D., & Woodruff, C. (2008). Returns to capital in microenterprises: Evidence from a
field experiment. Quarterly Journal of Economics, 123(4), 1329–1372.
Drexler, A., Fischer, G., & Schoar, A. (2014). Keeping It Simple: Financial Literacy and Rules of Thumb.
American Economic Journal: Applied Economics, 6(2), 1–31.
33
Dupas, P., & Robinson, J. (2013). Savings Constraints and Microenterprise Development: Evidence from
a Field Experiment in Kenya. American Economic Journal: Applied Economics, 5(1): 163–192.
Fairlie, R., Karlan, D., & Zinman, J. (2015). Behind the GATE Experiment: Evidence on Effects of and
Rationales for Subsidized Entrepreneurship Training. American Economic Journal: Economic Policy, 7(2), 125–
161.
Fiala, N. (2014). Stimulating Microenterprise Growth: Results from a Loans, Grants and Training
Experiment in Uganda. Working paper, German Institute for Economic Research, DIW Berlin.
Glaub, M., & Frese, M. (2011). A critical review of the effects of entrepreneurship training in developing
countries. Enterprise Development and Microfinance, 22(4).
Grimm, M., & Paffhausen, A-L. (2015). Do interventions targeted at micro-entrepreneurs and small and
medium-sized firms create jobs? A systematic review of the evidence for low and middle income
countries. Labour Economics, 32, 67–85.
Hsieh, C-T., & Klenow, P. J. (2009). Misallocation and Manufacturing TFP in China and India. Quarterly
Journal of Economics, 124(4), 1403-1446.
Karlan, D., & Valdivia, M. (2011). Teaching Entrepreneurship: Impact of Business Training on
Microfinance Clients and Institutions. The Review of Economics and Statistics, MIT Press, 93(2), 510-527.
Karlan, D., Knight, R., & Udry, C. (2012). Hoping to Win, Expected to Lose: Theory and Lessons on
Microenterprise Development. NBER Working Paper No. 18325.
Levinsohn, J., & Petrin, A. (2003). Estimating production functions using inputs to control for
unobservables. Rev. Econ. Stud., 70, 317–341.
Lindsay V., Ashill N., Roxas B., & Victorio A. (2014) Entrepreneurial orientation and performance of
microenterprises in an emerging economy. Journal of Strategic Marketing. 22(7): 631-656.
Lumpkin, G. T., & Dess, G. (1996). "Clarifying the Entrepreneurial Orientation Construct and Linking It
to Performance". Academy of Management Review, 21(1), 135–172.
McKenzie, D. (2011). How Can We Learn Whether Firm Policies Are Working in Africa? Challenges (and
Solutions?) For Experiments and Structural Models. Journal of African Economies, 20(4).
McKenzie D., & Woodruff C. (2014). What Are We Learning from Business Training and
Entrepreneurship Evaluations around the Developing World? World Bank Research Observer, 29 (1), 48-82.
McKenzie, D., & Woodruff, C. (2015). Business Practices in Small Firms in Developing Countries. Policy
Research Working Papers, 7405, the World Bank.
Miller, D. (1983). The correlates of entrepreneurship in three types of firms. Management Science. 29(7):
770–792.
Munoz, J.M., Welsh, D.H.B., Chan, S.H., & Raven, P.V. (2015). Microenterprises in Malaysia: a
preliminary study of the factors for management success. International Entrepreneurship and Management
Journal. 11:673–694.
Newmanan C., Rand J., Talbot T., & Tarp F. (2015). Technology transfers, foreign investment and
productivity spillovers. European Economic Review, 76, 168–187.
34
Olley, S., & Pakes, A. (1996). "The Dynamics of Productivity in the Telecommunications Equipment
Industry." Econometrica 64, 1263-98.
Petrin, A., Poi, B.P., & Levinsohn, J. (2004) Production function estimation in Stata using inputs to
control for unobservables. Stata Journal, 4(2), 113-123.
Rand, J. (2004). Documentation of the Small and Medium Scale Enterprise (SME) Survey in Vietnam for
the Year 2002. Hanoi, mimeo.
Rand, J., & Torm, N. (2012). The benefits of formalization: Evidence from Vietnamese manufacturing
SMEs. World Development, 40(5), 983–998.
Rand, J., & Tarp, F. (2011). Does Gender Influence the Provision of Fringe Benefits? Evidence From
Vietnamese SMEs. Feminist Economics, 17(1), 59-87.
Rand, J., & Tarp, F. (2012). Firm-Level Corruption in Vietnam. Economic Development and Cultural Change,
University of Chicago Press, 60(3), 571 - 595.
Rauch, A., Wiklund, J., Lumpkin, G.T., & Frese, M. (2009). Entrepreneurial orientation and business
performance: Cumulative empirical evidence. Entrepreneurship Theory and Practice. 33(3) : 761–788.
Sakai, H., & Takada, N. (2000). Developing Small and Medium Scale Enterprises in Vietnam, NRI Papers
No. 13, Nomura Research Institute.
Syverson, C. (2011). What determines productivity? Journal of Economic Literature, 49(2), 326–365.
van Beveren, I. (2010). Total sivity estimation: a practical review. J. Econ. Surv, 26(1), 98–128.
Wooldridge, ,J .M. (2009). On estimating firm-level production functions using proxy variables to control
for unobservables. Econ. Lett., 104, 112–114.
Yasar, M., Raciborski, R., & Poi, B.P. (2008). Production Function Estimation in Stata Using the Olley
and Pakes Method. Stata Journal, 8(2), 1–11.
35
Appendix
Appendix table 1. MCA results: contribution to the first dimension and adjusted inertia
Variable
Accounts
Advertise
Wage policy
Suppliers
location
Suppliers type
Outside services
0
1
0
1
None
Local comp.
Local SOE
Authorities
Agric rates
Negociations
Paying cap~y
Other
Dim.1
coordinates
0.236
-2.363
0.077
-2.644
2.677
-1.397
-1.812
-1.401
-1.242
-1.333
-1.303
-0.679
All commune
1.962
Province
-0.165
Country
Foreign
Private ent.
Only
Some HH.
Some SOE
0
1
-1.475
-8.426
Value
Variable
Pricing
Hiring method
Products
improvements
Innovation: products
0.25
0.177
-1.096
0.082
-2.378
Innovation: process
Innovation planning
Cost perc.
=Competitors
<Competitors
Bargain
Government
None
Newpaper
Labour exc~e
Friends
Authorities
Personnal rel.
Service centre
Dim.1
coordinates
-0.031
0.163
-0.28
0.061
-0.614
2.567
-1.746
-1.845
-1.43
-1.262
-1.207
-1.727
Other
-1.832
0
0.683
1
0
-1.507
0.053
1
-2.205
0
1
0
1
0.185
-2.433
0.56
-1.582
Value
Note: contributions based on MCA for all years. Negative values contribute positively to the MC index.
2007
2009
2011
2013
Dimension
dim 1
dim 2
dim 3
dim 4
60.38
65.77
70
72.04
59.19
66.49
69.63
71.33
Method: Burt/adjusted inertias
36
53.45
59.05
62.9
64.84
45.3
55.34
59.98
62.82
Appendix table 2. Marginal effects of MC on productivity by size category
Delta-method
dy/dx
Std. Err.
z
P>z
[95% Conf.
Interval]
1,2
0.093
0.018
5.060
0.000
0.057
0.129
]2,5]
0.053
0.010
5.110
0.000
0.032
0.073
]5,10]
0.041
0.018
2.280
0.023
0.006
0.077
11+
0.051
0.033
1.540
0.123
-0.014
0.116
37
Appendix table 3. MC and performance in the HB&ISS sample
MC index
Time
Log net profits
(2)
0.0785**
(0.0323)
0.300***
(0.0456)
(1)
0.0891***
(0.0327)
0.334***
(0.0407)
Interactions:
Gender
0.00467
(0.0693)
-0.0527
(0.0683)
Youth
Informal (no BRC)
Size (# workers)
Invested in the past year
-0.0895
(0.0999)
0.254**
(0.109)
-0.0473
(0.0817)
9.838***
(0.160)
-0.0398
(0.0930)
0.229**
(0.103)
-0.0545
(0.0816)
-0.0739
(0.125)
-0.0182
(0.133)
0.0244
(0.0831)
0.112
(0.106)
0.132*
(0.0713)
0.389
(0.240)
9.756***
(0.230)
4,016
0.040
2,008
4,016
0.046
2,008
Premise: at home
Premise: dedicated
Access to water
Access to phone
Access to mobile phone
Access to internet
Constant
Observations
R-squared
Number of id
(3)
38